Result: no detectable difference, so far
So far, we don't see fast-growing business demand pushing household bills above similar areas.
We compared 85 utility service areas in 31 states (65 companies; a company serving several states counts once per state, because prices are set state by state). The third whose business customers' electricity use grew fastest from 2019 to 2022 were set against similar utilities in the same region, and we looked at what happened to household electricity prices from January 2024 to June 2026. After allowing for gas prices and for changes that hit the whole region, the fast-growth group's prices did not rise clearly more than their neighbours'. The plausible range runs from about $7 less a month to about $3 more a month for a typical home. Our best estimate is about $2 a month less, but chance alone produces a gap that size about 40 times in a hundred, so we don't read anything into the direction.
This compares groups of utilities. It says nothing about any one household's bill.
"Bills" here means the average residential price per kWh (all residential revenue divided by all residential sales, EIA-861M), turned into dollars for a home using 899 kWh a month at the fast-growth group's average 2023 price of 13.7 cents per kWh. A real bill also depends on how much you use. Prices are nominal, through June 2026; recent months are preliminary.
This is not proof that growing demand has no effect on bills. Our test only asks whether the fastest-growing areas pulled away from their own neighbours. If demand pushed prices up across a whole region, both groups would rise together and this test wouldn't show it. It also leaves out PJM (including Virginia), where data-centre growth has been most discussed, and it only runs to mid-2026. Utilities usually recover the cost of new plants and power lines through rate cases that take time, so costs that haven't reached bills yet can't show up here.
- Neighbours only. A rise that hit a whole region at once is invisible to this test.
- Not the places in the news. PJM, including northern Virginia, is not in the test.
- Too early. The data stop in June 2026, and rate cases reach bills with a delay.
- Price, not a bill. We measured price per kWh and converted it for a typical home. A household's usage, plan and supplier all change what it actually pays.
The main test covers investor-owned utilities with at least ten thousand homes that sell both the power and the delivery, in the four regions that had enough such utilities to compare: the MISO and SPP grids, and the rest of the South and West.
It leaves out all of PJM (the grid covering Virginia, Maryland, Pennsylvania, Ohio, New Jersey and neighbours, including northern Virginia), Texas's ERCOT, New York, New England, California's grid operator, Alaska and Hawaii. Most of those sell power and delivery separately; the PJM utilities that sell both were too few to test as a region. If a household's utility is not in the test, that is a limit of the method, not a sign that nothing is happening there.
The range next to the smallest gap we planned to see
The bar shows the range of gaps that fit our data. The dashed lines show how big a gap would have to be for our method to reliably notice it, as calculated before we ran the test. Smaller gaps can slip past. Anything under about $5 a month is hard for this test to tell apart from zero.
How we got the range: we reshuffled which utilities counted as "fast-growth" 9,999 times, only swapping within the same region, and measured the gap each time. Chance alone produced a gap at least this large about 40 times in a hundred. That is common, so we do not read the gap as real.
Before 2024: were the two groups already drifting apart?
This is the check that decides whether the comparison is fair. If the fast-growth utilities had already been pulling away from their neighbours before 2024, we could not interpret anything after. They had not: every yearly gap is small and every bar includes zero. On the right is the single pre-registered estimate for 2024 to June 2026 taken together, with its range, which also includes zero.
We do not show year-by-year estimates for 2024 onward. They were not part of the plan, 2026 is only six months of preliminary data, and any one year would be easy to over-read.
Show this chart as a table
| Year | Best guess | 95% range |
|---|---|---|
| 2015 | +0.5% | -4.4% to +5.7% |
| 2016 | -0.9% | -6.4% to +4.8% |
| 2017 | -2.0% | -7.2% to +3.6% |
| 2018 | -1.4% | -6.1% to +3.6% |
| 2019 | -0.7% | -5.3% to +4.2% |
| 2020 | -1.2% | -5.8% to +3.6% |
| 2021 | -1.3% | -6.2% to +3.8% |
| 2022 | -1.3% | -5.6% to +3.1% |
| 2023 | reference year (zero by definition) | |
| 2024 to Jun 2026, together | -1.7% | -5.5% to +2.2% |
What we can't say
- We can't say anything about any one household's bill. This compares groups. A household's usage, plan, supplier and state rules all matter more to its bill than anything here.
- We can't say it's data centres, or that it isn't. We measured fast-growing business power demand, which includes factories, electrification and the rebound after COVID. Our check of how well this measure picks out data-centre areas came out only partly, so we do not name data centres in the claim.
- We can't see a region-wide rise. If growing demand raised prices for everyone in a grid region, both groups would rise together and the gap would stay near zero.
- We can't speak for the places we left out. That includes the Mid-Atlantic, including northern Virginia.
- We can't see small effects. Anything under about $5 a month is hard to tell apart from zero. Before running the test we worked out that a rise of about $2 a month would have been caught about 24 times in a hundred.
- We can't see the future. This is the past through June 2026. The measure looks backward too: growth was measured only up to 2022, so places whose large new loads arrived later sit in the comparison group, which would make a real gap look smaller.
- We can't say why prices moved. Comparing within a region removes what hit everyone alike. It does not remove everything that hit one place.
- If the groups would have drifted apart somewhat anyway, the range widens to roughly $8 less to $5 more a month. This is a simplified side check and does not change the headline.
Other things that changed over the same period
Gas prices are one of the obvious suspects when bills move, because natural gas sets much of the price of power. Below is the wholesale price over the same years. We don't claim any of the things on this page explains the result.
Show this chart as a table
| Year | Average $/MMBtu |
|---|---|
| 2015 | $2.63 |
| 2016 | $2.52 |
| 2017 | $2.99 |
| 2018 | $3.17 |
| 2019 | $2.57 |
| 2020 | $2.03 |
| 2021 | $3.91 |
| 2022 | $6.42 |
| 2023 | $2.54 |
| 2024 | $2.19 |
| 2025 | $3.53 |
| 2026 (to Aug) | $3.61 |
Not charted in this first release: regional inflation, and wildfire and storm-related rate requests. We have not built those charts, so the list of other changes here is incomplete.
Why a pattern isn't a cause
Two things can move together without one causing the other. If bills rose faster where a new data centre opened, that could be the data centre, or a storm, or an old power plant retiring, or the price of gas. We compare against similar places and subtract what everyone saw, which helps. It does not make the comparison perfect.
The reverse also holds. Because we did not find a gap, you cannot conclude that gas prices or anything else is the real reason for a bill. A null result leaves the other explanations where they were.
Looking up your own utility
This first release has no ZIP or utility lookup, on purpose. A fair one needs a map from ZIP code to the utility that serves it, and we do not have that data yet. Many ZIP codes are served by more than one utility, so guessing would sometimes show the wrong one. A per-utility page would also need its own review, so that nothing reads as a verdict on a named company. When it exists, your ZIP will stay in your browser and will not be sent anywhere.
Check our work
We wrote down the exact test, and how we would report every possible result, before running it, and fixed a fingerprint of that plan with an outside timestamping service. How to check it yourself, and what we did differently from the plan. How we tested this, in a few paragraphs.